Detect emerging, bridge-candidate, and undercited papers inside a curated audio-ML corpus, then expose the signals behind every recommendation.
Audio ML paper discovery
Research Radar is a working prototype for ranking and explaining MIR + audio representation learning papers. It keeps the corpus narrow, shows the ranking signals, and explains why each paper appears.
Product frame
Neural audio effects and music generation stay as a controlled edge set so bridge-paper logic remains meaningful and the recommendation surface stays coherent.
Rank papers by citation velocity and local topic growth, with semantic fields shown only when a selected run computes and uses them.
Inspect cross-cluster candidates and the signal evidence recorded for the selected run. Bridge is experimental and separate from the default recommendation path.
Expose score contributions, cluster context, and baseline comparisons so each recommendation feels engineered.
Pair baseline comparison with simple citation and date sorts so ranking behavior can be inspected before stronger quality claims are made.
Product promise
Search hands off into ranked views, topic momentum, and evaluation pages, so discovery work does not end at a single results screen.
The bridge view shows cluster context and recorded signal evidence without presenting the signal as a validated recommender.
Ranking weights, snapshot versions, and topic signals should read like instrument output, not hidden implementation detail.